My question would be: how deterministic is this algorithm? On a large corpus the program outputs something in the "style" of the composer. (This happens to human composers too, sometimes: late in life they start turning out self-parodies.) But given a single piece, can the EMI algorithms (without jiggering) produce the same piece, or a piece indistinguishable from it? Unless it could do that, I don't see in what way it replaces or replicates the composer.